Systems and methods for decoding of graph-based channel codes via reinforcement learning

Abstract

Embodiments of the present disclosure relate to sequential decoding of moderate length low-density parity-check (LDPC) codes via reinforcement learning (RL). The sequential decoding scheme is modeled as a Markov decision process (MDP), and an optimized cluster scheduling policy is subsequently obtained via RL. A software agent is trained to schedule all check nodes (CNs) in a cluster, and all clusters in every iteration. A new RL state space model is provided that enables the RL-based decoder to be suitable for longer LDPC codes.

The invention claimed is: 
     
       1. A method for decoding low-density parity-check codes encoded in a traffic channel of a communication signal received by a mobile communication device, the method comprising:
 generating a decoding schedule for each of a plurality of clusters of check nodes in response to execution of a reinforcement learning software agent of a low-density parity check (LDPC) decoder, wherein the decoding schedule is generated based on a reward associated with an outcome of decoding each of the plurality of clusters of check nodes, and wherein the reward corresponds to a probability that corrupted bits of the communication signal are correctly reconstructed; 
 sequentially decoding each of the plurality of clusters of check nodes according to the decoding schedule; 
 updating a posterior log-likelihood ratio of all variable nodes (VNs) based on the sequential decoding schedule; 
 determining whether a specified maximum number of iterations has been reached or a stopping condition has been satisfied based on the sequential decoding schedule; 
 in response to determining that the specified maximum number of iterations is reached or the stopping condition is satisfied, outputting a reconstructed signal corresponding to the communication signal received by the mobile communication device. 
 
     
     
       2. The method of  claim 1 , further comprising establishing a cluster scheduling policy based on the training. 
     
     
       3. The method of  claim 2 , wherein the decoding schedule is determined based on the cluster scheduling policy. 
     
     
       4. The method of  claim 1 , further comprising clustering the check nodes into the plurality of clusters to minimize inter-cluster dependency. 
     
     
       5. The method of  claim 1 , wherein the reinforcement learning software agent implements at least one of a Q-learning or a deep reinforcement learning scheme to generate the cluster scheduling policy. 
     
     
       6. A system for decoding low-density parity-check codes encoded in a traffic channel of a communication signal received by a mobile communication device, the system comprising:
 a non-transitory computer-readable medium storing instructions for decoding low-density parity-check codes; and 
 a processing device executing the instructions to:
 generate a decoding schedule for each of a plurality of clusters of check nodes in response to execution of a reinforcement learning software agent of a low-density parity check (LDPC) decoder, wherein the decoding schedule is generated based on a reward associated with an outcome of decoding each of the plurality of clusters of check nodes, and wherein the reward corresponds to a probability that corrupted bits of the communication signal are correctly reconstructed; 
 sequentially decode each of the plurality of clusters of check nodes according to the learned scheduling policy; 
 update a posterior log-likelihood ratio of all variable nodes (VNs) based on the sequential decoding schedule; 
 determine whether a specified maximum number of iterations has been reached or a stopping condition has been satisfied based on the sequential scheduling policy; 
 output a reconstructed signal corresponding to the communication signal received by the mobile communication device in response to determining that the specified maximum number of iterations is reached, or the stopping condition is satisfied. 
 
 
     
     
       7. The system of  claim 6 , wherein the processing device executes the instructions to establish a cluster scheduling policy based on the training. 
     
     
       8. The system of  claim 7 , wherein the decoding schedule is determined based on the learned cluster scheduling policy. 
     
     
       9. The system of  claim 6 , wherein the processing device executes the instructions to cluster the check nodes into the plurality of clusters to minimize inter-cluster dependency. 
     
     
       10. The system of  claim 6 , wherein the reinforcement learning software agent implements at least one of a Q-learning or a deep reinforcement learning to generate the decoding schedule. 
     
     
       11. A non-transitory computer-readable medium comprising instructions, wherein execution of the instructions by a processing device causes the processing device to:
 generate a decoding schedule for each of a plurality of clusters of check nodes in response to execution of a reinforcement learning software agent of a low-density parity check (LDPC) decoder, wherein the decoding schedule is generated based on a reward associated with an outcome of decoding each of the plurality of clusters of check nodes, and wherein the reward corresponds to a probability that corrupted bits of the communication signal are correctly reconstructed; 
 sequentially decode each of the plurality of clusters of check nodes according to the learned scheduling policy; 
 update a posterior log-likelihood ratio of all variable nodes (VNs) based on the learned sequential scheduling policy; 
 determine whether a specified maximum number of iterations has been reached or a stopping condition has been satisfied based on the sequential cluster scheduling policy; 
 output a reconstructed signal corresponding to the communication signal received by the mobile communication device in response to determining that the specified maximum number of iterations is reached or the stopping condition is satisfied. 
 
     
     
       12. The medium of  claim 11 , wherein execution of the instructions by the processing device causes the processing device to establish a cluster scheduling policy based the training. 
     
     
       13. The medium of  claim 12 , wherein the decoding schedule is determined based on the sequential cluster scheduling policy. 
     
     
       14. The medium of  claim 11 , wherein the reinforcement learning software agent implements at least one of a Q-learning or a deep reinforcement learning to generate the decoding schedule.